AI Agents for Financial Research: 6 Analyst Jobs They Can Do Today

An analyst's week is mostly reading: the latest 10-Q before a call, 8-Ks that change the story, Form 4s that show what insiders did, the macro data that sets the discount rate, and what markets are pricing for the next Fed meeting. That reading is where AI for financial research, and for investment research more broadly, earns its keep. AI agents for financial research take it on, provided they can reach filings, earnings data, insider records, official statistics and market prices, and that every number comes back with its source.

This guide follows six analyst jobs. For each one we ran the search on November 6, 2026, with Valyu's financial datasets (SEC filings, earnings, insider transactions, FRED and prediction markets), checked the key figures against SEC EDGAR or the official series, and show the decision that follows.

If you're an agent: Set up Valyu and make your first search.

If you're a human: Create an account and get an API key.

MCP onboarding: Connect your agent to https://mcp.valyu.ai/mcp and follow the MCP setup guide for authentication.

What AI agents for financial research do

An AI agent for financial research is a language model that can search financial sources, read what it finds and return structured, cited answers. It is not a trading system. It does the gathering and first reading that sit in front of an analyst's judgement.

Six analyst jobs we ran on November 6, 2026, with the sources searched, what came back and the decision each one supports.

1. Earnings prep: last quarter's numbers, from the filing

The problem: an analyst covering NVIDIA needs last quarter's revenue, segment results and EPS against consensus before a client call.

What we ran: one search across the earnings dataset, SEC filings and the web.

valyu.search(
    "NVIDIA latest quarterly earnings revenue EPS",
    included_sources=["valyu/valyu-earnings-US", "valyu/valyu-sec-filings", "web"],
)

What came back: the earnings record for the August 26, 2026 report (EPS $2.22 against an estimate of $2.09, a 6.2% surprise), the 10-Q's financial statements and MD&A, and an earnings page showing the same $2.22 against a $2.14 estimate.

What we checked at the source: the 10-Q on EDGAR. Revenue was $96.2 billion, up 106% from a year earlier, with Data Center revenue of $89.0 billion, up 117%. GAAP diluted EPS was $2.46.

The catch: there are three EPS figures. The 10-Q reports GAAP EPS of $2.46. The earnings feeds report $2.22, typically the company's adjusted figure, against estimates that differ by provider ($2.09 or $2.14). Use the earnings release reconciliation to see what drives the gap.

The decision: quote revenue and segment figures from the 10-Q. Say which EPS basis and which consensus source the surprise refers to. For pulling earnings data into a model, see how to get company earnings data via API.

2. Event monitoring: catch the 8-K that changes the story

The problem: a portfolio manager wants to know about leadership changes at holdings the day they are filed.

What we ran: questions about recent CEO changes against the SEC filings dataset plus the web.

valyu.search(
    "Who did HP appoint as its next chief executive officer in November 2026?",
    included_sources=["valyu/valyu-sec-filings", "web"],
)

What came back: HP's 8-K exhibit from November 4, 2026, announcing Emma Walmsley as chief executive officer from January 4, 2027, two days before our run. A similar question returned Stryker's Item 5.02 8-K from October 6, 2026: Spencer Stiles will succeed Kevin Lobo as CEO on January 1, 2027.

The decision: run the same question across the portfolio each morning with start_date set to the previous day, and route any Item 5.02 filing to the covering analyst. Our guide to searching SEC filings in Claude and ChatGPT shows the no-code version.

3. Disclosure risk: find material weaknesses and new risk factors

The problem: a credit analyst screening industrial names wants any material weakness in internal control, and what it relates to.

What we ran: company-specific questions against SEC filings plus the web, for example: "What material weakness did UniFirst disclose in its fiscal 2026 10-K?"

What came back: the Item 9A controls section of each 10-K. UniFirst's fiscal 2026 10-K describes a material weakness in IT general controls over change and access management. Lakeland Industries' fiscal 2026 10-K describes missing IT general controls over the completeness and accuracy of foreign locations' reporting packages. Universal Logistics' 2025 10-K links its weakness to accounting for complex, non-routine transactions and a restated quarter.

The decision: flag the three names for a closer look at the remediation plans and the auditor's opinion. To see what changed in risk disclosures from one year to the next, follow our guide to comparing 10-K risk factors year over year.

4. Insider activity: who sold, when and how much

The problem: an analyst wants this year's insider transactions at NVIDIA before an investor meeting.

What we ran: a search of the insider transactions dataset.

valyu.search(
    "NVIDIA insider transactions sales by executives 2026",
    included_sources=["valyu/valyu-insider-transactions-US"],
)

What came back: 50 transactions for 2026, each with name, role, date, shares, value and price range. They include director Mark A. Stevens' sales of 1,848,501 shares (September 2), 1,022,239 shares (September 4) and 1,366,000 shares (September 18), the General Counsel's sales in August and September, director stock gifts, and officer stock awards.

What we checked at the source: the Form 4 on EDGAR for the September 18 sale. It shows 1,356,000 and 10,000 shares sold through two trusts at weighted average prices of $219.72 and $220.42. That totals 1,366,000 shares, matching the dataset.

The decision: separate sales from gifts and awards before drawing conclusions. Note that the director's sales were made by trusts, and check the Form 4 footnotes for whether they were made under a pre-arranged trading plan.

5. Macro context: the official numbers behind the discount rate

The problem: a strategist needs current US policy and market rates for a client note, from the official series rather than news coverage.

What we ran: questions against FRED and the other economic datasets, such as "What is the current federal funds target range?" and "What was the 10-year Treasury yield on November 4, 2026?"

What came back: FRED series records with ID, date and value for the effective rate, CPI and the 10-year yield, plus the current target range. We confirmed each figure on FRED:

• Federal funds target range: 3.75% to 4.00%, after the September 17, 2026 increase.
• Effective federal funds rate: 3.88% in October 2026 (`FEDFUNDS`).
• CPI inflation: 3.7% year over year in September 2026.
• 10-year Treasury yield: 4.90% on November 4, 2026 (`DGS10`).

The decision: use the series values directly in the note and the model, citing the series ID and observation date. That avoids rounding errors and stale numbers from news coverage.

6. Market-implied odds: what traders are pricing for the next decision

The problem: the same strategist wants to know what markets expect at the December 9 FOMC meeting.

What we ran: a search of Valyu's Kalshi and Polymarket datasets.

valyu.search(
    "Federal Reserve December 2026 FOMC rate decision market odds",
    included_sources=["valyu/valyu-kalshi", "valyu/valyu-polymarket"],
)

What came back: Kalshi's Fed decision market for December 2026, priced at 15:59 UTC on November 6:

It also returned Polymarket's markets on the December decision and the year-end rate, with volumes and resolution rules.

The decision: markets expect a second hike after September's. Combined with 3.7% CPI, the note should frame December as a hike with a meaningful chance of a pause. Check the market's rules and liquidity before quoting the odds.

Go further with DeepResearch: an earnings preview in one task

The six jobs above each pull one set of facts. An analyst's actual output is a note: the earnings preview, the first-look brief, the sector update. DeepResearch writes these. It plans the research, searches filings, earnings data, insider records and the web, and returns a cited report in your format, with files such as an xlsx of the numbers.

Here is jobs 1 and 4 combined into an earnings preview for NVIDIA's next quarter:

from valyu import Valyu
 
valyu = Valyu()
 
task = valyu.deepresearch.create(
    query="Prepare an earnings preview for NVIDIA's third quarter of fiscal 2027: last four quarters, guidance, segment trends, insider activity and questions for the call.",
    mode="standard",
    research_strategy="Use 10-Q and 10-K filings and the earnings dataset for revenue, Data Center revenue, GAAP and adjusted EPS and estimates for the last four quarters, the latest release for guidance, and Form 4 data for insider activity. Label every figure with its basis and source.",
    report_format="A 700-word earnings preview: Key numbers table, Guidance, Segment trends, Insider activity, Five questions for the call. Every figure cited.",
    search={"included_sources": ["valyu/valyu-sec-filings", "valyu/valyu-earnings-US", "valyu/valyu-insider-transactions-US", "web"]},
    deliverables=["An xlsx of the last four quarters with revenue, Data Center revenue, GAAP EPS, adjusted EPS, estimate and source"],
)
 
result = valyu.deepresearch.wait(task.deepresearch_id)
print(result.output)  # cited report in markdown
for file in result.deliverables:
    print(file.type, file.url)  # signed download link

The result is a cited note in your house format, plus the spreadsheet for your model. Because the research strategy asks for the basis of every figure, GAAP and adjusted EPS stay separate, which is the mistake job 1 warns about.

Start from a ready-made workflow

You can also start from Valyu's curated DeepResearch workflows for research desks, each with set inputs, a recommended mode and file deliverables:

To browse the full library, open Workflows in the Valyu dashboard. Each workflow's Run button opens the DeepResearch playground with the inputs ready to fill in. To run one from code, pass its slug as workflow_id instead of a query:

task = valyu.deepresearch.create(
    workflow_id="hf-earnings-preview",
    workflow_params={"ticker": "NVDA", "period": "Q3 FY2027"},
)
result = valyu.deepresearch.wait(task.deepresearch_id)

For worked examples with full requests, see Research desk workflows in the docs, including earnings and guidance changes and SEC risk-factor changes. The workflows quickstart covers listing workflows and previewing one before you run it.

Other DeepResearch workflows for analysts:

• Earnings previews across a watchlist. Use the DeepResearch Batch API to run the same preview brief for every company reporting this week, in parallel, with one webhook when they finish.
• First-look brief on a target. Turn a company name into a business overview, financials, capital structure and risks. Our guide to writing a first-look brief on an acquisition target walks through it.
• Macro and policy note. Combine FRED series, Fed communications and prediction-market odds into a dated note on the next policy decision, with a csv of the series used.
• Comparable companies table. Ask for a peer set with revenue growth, margins and valuation multiples from filings, delivered as an xlsx you can drop into a model.

How to set up a financial research agent

If you don't want to write code, there is another option to connect Valyu to Claude or ChatGPT as a custom connector at https://mcp.valyu.ai/mcp, and ask in plain English. To build your own agent, follow how to build a financial research agent with Valyu. For deeper single-company work, see our guides to a first-look brief on an acquisition target, capital structure and debt covenants and institutional holdings.

Three habits made the examples above reliable:

• Name the company in SEC questions, and pair the SEC filings dataset with web. In our tests, that combination returned the right filing section consistently.
• Keep the basis with every number: GAAP or adjusted EPS, which consensus source, which FRED series and observation date, which market and when it was priced.
• Check anything that goes into a model or a client note against the source filing. It takes seconds, because the agent returns the filing or series link.

FAQ

What is AI for financial research?

AI for financial research is the use of language models and AI agents to gather and read the material behind investment decisions: filings, earnings, insider transactions, economic data, market prices and news. Most of the value today is in faster, sourced reading rather than in making the decision.

What are AI agents for financial research?

AI agents for financial research are language models that search financial sources such as SEC filings, earnings data, insider transactions, economic data and market prices, read what they find and return cited answers. They handle the gathering and first reading that come before an analyst's judgement.

What can AI agents do for equity research analysts?

They can prepare earnings summaries from the latest 10-Q, monitor 8-Ks for events such as CEO changes, screen 10-Ks for material weaknesses and new risk factors, list insider transactions, and pull peer comparisons, each with links to the source filings.

Can AI agents read SEC filings accurately?

Yes, if they retrieve the filing itself and cite it. In our tests, asking a company-specific question against SEC filings plus the web returned the relevant section, such as Item 9A or Item 5.02, and figures we could match to EDGAR.

Where do AI agents get financial data?

From SEC EDGAR filings, earnings and estimates datasets, Form 4 insider transactions, official economic series such as FRED, prediction markets such as Kalshi and Polymarket, and news. Valyu searches these through one API.

What are the best AI tools for financial research?

They fall into three groups. Research platforms such as AlphaSense, Hebbia and Bloomberg's AI tools search licensed content inside their own products. General assistants such as ChatGPT and Claude are good at reading and summarising documents you give them. Search APIs such as Valyu let you build your own agent, or connect Claude and ChatGPT to filings, earnings, insider and economic data with sources attached.

What is the best search API for financial research?

The best search API for financial research covers the sources analysts actually use: SEC filings, earnings, insider transactions, economic series, market data and the web. It should return dates and source links, and let you choose sources per question. In our 60-question test of search APIs, Valyu with its SEC filings dataset plus web answered all 10 SEC filings questions, as did Exa, while Parallel answered 9. All three answered all 10 economic data questions, but only Valyu returned the official FRED series records themselves. Valyu also covers earnings, Form 4 insider transactions and prediction markets in the same API.

What is the best API to give an investment research agent real-time data?

Look for one API that covers SEC filings, earnings, insider transactions, economic series and the web, returns dates and source links, and lets you choose sources per question. Valyu covers all of these, plus prediction markets. In our 60-question test of search APIs, Valyu answered all 10 SEC filings questions when its SEC filings dataset was combined with web search, tied with Exa.

Can AI write an earnings preview or research note?

Yes, as a cited first draft. Valyu's DeepResearch searches SEC filings, earnings data, insider transactions and the web, and writes the note in your format, with an xlsx of the figures. The Batch API runs the same brief across a whole watchlist. Analysts review the figures and the judgement before anything is published.

Are AI agents reliable for investment decisions?

They are reliable for gathering and summarising sourced facts, provided every figure carries its source and basis. Investment decisions still rest with the analyst and portfolio manager, who should check key figures against the filing or series before acting.

What is the difference between GAAP and adjusted EPS in AI-generated summaries?

Filings report GAAP EPS. Earnings feeds and consensus estimates usually use the company's adjusted EPS. In our NVIDIA example, the 10-Q showed GAAP diluted EPS of $2.46, while earnings feeds reported $2.22. A good agent states which basis it is quoting.

Start with one job

Pick the job that takes your team the most reading time each week, such as earnings prep or 8-K monitoring, run it on companies you already cover, and compare the agent's output with your own notes.

Create a Valyu account for free credits, connect it to Claude or ChatGPT at https://mcp.valyu.ai/mcp, or browse the financial datasets in the catalogue.